Guidelines for building fixincludes for targets?

[E. Weddington]

Hello!

I have some questions about the guidelines for fixincludes and cross 
toolsets. To refresh, I build a toolset for --target=avr --host=mingw32, 
and fixincl (and as a side note, collect2) cannot be built for this 
combination (mainly because of host=mingw32).

Originally I had problems building collect2 for mingw32 as referenced in 
bug #14316:
<http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14316>

In the bug comments, fixincludes was also discussed, and there seemed to 
be some discussion between Zack and Ian as to whether fixincludes should 
be built for a cross toolset. It looks like it ended up that fixincl 
should be built even if a cross is being built. Also, comment #14 (by 
Ian) was a great explanation on what fixinc is supposed to do. [As 
another side note, bug #14316 *still* needs the patch committed.]

So I submitted bug #17462, which has now been rolled into bug #17832:
<http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17832>
Aaron seems to be taking care of getting fixincl built for host=mingw32 
(Thanks, Aaron!)

But, regarding the avr target specifically, there should be no header 
files for the avr target to fix. The C library that is used for the avr 
target is avr-libc: <http://savannah.nongnu.org/projects/avr-libc/>, and 
is built specifically to work with binutils/gcc. This is the C library 
that is used to build the avr toolset for Linux, FreeBSD, Windows, Mac 
OSX, and Solaris.

So, I've been building the toolset with a one line patch to 
gcc/fixinc/mkfixinc.sh that adds the avr-*-* triplet to the list of 
targets to not build fixincl for. This has worked fine for me (and many 
others) for over a year.

1. Is mkfixinc.sh the correct place to do this?
2. Would not building fixincl make sense for this target?
3. I assume that I should do the usual route of submitting bug report, 
patch to gcc-patches, etc. Correct?

Thanks
Eric